1. It is crucial to avoid overexposing the highlights in color checker pictures. That was my main problem. You can use the Raw Histogram of ART/RT to check the highlights. Take a lot of pictures with different exposures and then select the best one.
2. In real life, it is often sufficient to use a single D50 illuminant color profile.
3. On my Linux system, argyll-scanin is called scanin (installed from the argyll package).
4. The path to the ColorChecher.cht file is also different: /usr/share/color/argyll/ref/ColorChecker.cht
5. X-Rite/Calibrate changed their formula in ColorCheckers produced November 2014 and later. cc24_ref.cie (from DCamProf) is for targets produced before November 2014. While cc24_ref-new.cie is for targets produced November 2014 and later. Therefore, I use the cc24_ref-new.cie file.
6. You can try the -s option with make-profile to use an alternate optimization algorithm.
7. Pay attention to the DE values in the make-profile result report. In my experience, they should not exceed 5. If the values are too high (10+), take another picture of the color checker.
8. It is better not to use an embedded tone curve for ART/RT. So I use the -t none option with make-dcp, and disable the "Look table" in ART/RT.
9. I also prefer to completely remove the "Look table" from a dcp file. I use the dcp2json and json2dcp commands of DCamProg; in the intermediate JSON, I remove the "ProfileLookTable" field.
10. Do not use the HueSatMap 3D (the -D option with make-dcp). In some cases, it can worsen highlight recovery in ART/RT.
